Monday, September 30, 2024

Improving the HRVT1 agreement - IJSPP 10/2024

Much has been written and discussed regarding the agreement between the first HRV a1 threshold (a DFA a1 of 0.75) and established standards (VT1, GET, LT1). There are multiple studies and opinions both for and against whether this marker is a valid proxy of the first ventilatory or lactate threshold. Who is correct, are any correct?  

After the last a1 article that appeared in MSSE, I felt that a "back to the drawing board" approach was needed to better answer this question. After some pondering, I decided not to discard our original concept of the HRVT1, that of being midway between a "correlated" and an uncorrelated DFA a1 pattern. Instead, after looking at the participant ramp data done over multiple studies by different groups, it appeared that the "midpoint" was not always at an a1 of 0.75. so a mathematical correction was needed. With the help of my co-authors and a set of blinded RR data, an attempt was made at estimating the HRVT1 based on individual midpoints for each person's ramp. The results came out fairly well, and in this post I will provide some details from the published article. References are also included.

Introduction
The identification of physiological boundaries pertaining to endurance exercise is of fundamental importance for both fitness monitoring and training intensity distribution 1,2. Unfortunately, proven methods to obtain exercise intensity thresholds separating exercise intensity domains (i.e., moderate to heavy, heavy to severe domains) often require elaborate equipment for evaluation of gas exchange and ventilatory responses, or invasive monitoring such as lactate testing. Therefore, the search for “surrogate” markers of these exercise boundaries without these requirements has attracted much interest. One such concept revolves around the measurement of a nonlinear index of heart rate (HR) variability (HRV) termed Detrended Fluctuation Analysis alpha 1 (DFA a1) during ramp incremental testing 3. DFA a1 is related to the degree of fractal behavior of the cardiac beat sequence as well as “correlation properties” over short time measurements 4,5. Like coastlines, snowflakes or tree branching, fractals are regarded as structures that possess self-similarity at various degrees of magnification. No matter the scale or magnification, the fundamental pattern is similar. As opposed to these structural examples, fractal behavior of the cardiac beat sequence is characterized by degrees of self-similarity of the beat series over different time scales6. This fractal behavior can also be mathematically quantified as “correlation properties” of the cardiac beat repetition patterns over variable time spans. To better comprehend the concept of correlation properties, parallels to a random walk have been drawn5. For example, during a random walk, at each next step, the walker can choose to go either right or left. If the choice the walker makes is not random but based on the previous sequence (series of right or left decisions), the pattern is described as being well “correlated” (DFA a1 near or above 1.0), since the future pattern is based on the past history. Values above 1.0 denote progressively higher degrees of correlation. But, if each new step is taken with equal, random chances of right or left, an “uncorrelated” pattern exists (DFA a1 of 0.5).Unlike most conventional HRV indexes that change with exercise, DFA a1 has the advantage of a wider dynamic range encompassing the moderate, heavy, and severe exercise intensity domains 7–9.
For these surrogate thresholds to be of realistic use, good agreement to existing standards as well as day to day repeatability needs to be present. Studies by multiple groups using both cycling and treadmill models have shown encouraging results supporting the second HRV threshold (HRVT2, DFA a1 value of 0.5) as a viable marker of the respiratory compensation point (RCP) or second lactate threshold 10–15. Additionally, DFA a1 behavior at constant power 16, during replicate ramp incremental testing 14,15 and over differing ramp incremental slope protocols 17 has been shown to be reliable/repeatable. However, in the case of the first HRV threshold (DFA a1 value of 0.75) agreement with the gas exchange threshold (GET) or first lactate threshold, results are somewhat conflictive. Although HRVT1 has been demonstrated to have strong correlation and minimal bias to different markers of the first threshold during incremental tests on the treadmill 18,19 or during cycling exercise 11,20, other studies have not shown such agreement. For example, a report comparing different ramp incremental slopes 17, three studies involving cycling 12,14,15 and one composed of only female participants 21 have been consistent with an overestimation of the first HRV threshold in comparison to traditional markers of the first threshold. However, even though the first HRV threshold was overestimated in some of these these trials, the reliability/repeatability was still good 14,15. Another situation that could theoretically lead to HRV threshold divergence is that of participant fitness status. However, this does not seem relevant as studies with high athletic fitness show both good agreement11,19,20 and poor agreement14 with GET or first lactate thresholds. It should be noted that DFA a1 including the first and second HRV thresholds can be affected by HRV artifact correction, recording device, sensor lead placement and preprocessing software 3,22 which could potentially affect results. However, careful examination of studies evaluating the first HRV threshold to gas exchange or lactate-derived indicators generally employs similar methodology, artifact correction and are even authored by similar research groups.
To explore why these differences in the first HRV threshold to the GET or the lactate threshold agreement may occur, an examination of the reasoning behind the HRVT1 concept is in order. DFA a1 correlation patterns are thought to be due to changes in sinoatrial pacemaker function controlled by the balance between the branches of the autonomic nervous system (ANS) 23. During exercise there is both a withdrawal of parasympathetic and an increase of sympathetic activity resulting in a change of HRV including DFA a1 24. Past studies demonstrated that during a ramp incremental test, values were well above 1.0 at very low intensities (well correlated), moved through a “partially” correlated (±0.75) region at moderate intensities, passed the “uncorrelated” value of 0.5 near the heavy/severe intensity boundary, finally reaching values below 0.5, signifying an “anticorrelated” pattern at severe intensities 25. It was posited that at the first threshold, the cardiac beat pattern would be found in an intermediate zone between well correlated (DFA a1 ≥ 1) and uncorrelated (DFA a1 = 0.5) behavior, which was numerically set to 0.75 7. In addition, it was conjectured that the second threshold would correspond to a second DFA a1 threshold of 0.5 (HRVT2), which represents a random interbeat pattern 10. DFA a1 values below 0.5 correspond to “anticorrelated” patterns representing an autonomic response indicating organismic destabilization consistent with exceeding a metabolic steady state 5,26. More careful inspection of prior data indicates that some individuals have a DFA a1 well above 1.0 early in the ramp incremental, with these values still considered strongly correlated 27. Since the initial DFA a1 in some individuals may be higher, the midpoint between well correlated and uncorrelated values (defining the first HRV threshold) in these cases would be shifted to a higher DFA a1 as a mathematical correction. For example, if an individual had a DFA a1 of 1.5 early in the ramp incremental, the midpoint denoting the first HRV threshold would be 1.0 with the associated HR and V̇O2 calculated accordingly. This suggests that the lack of agreement that some studies observed between the physiological responses observed at the first threshold might be explained, at least to a given extent, by the methodological approach used for the first HRV threshold determination.
Therefore, the aim of this study was to compare the agreement between the V̇O2/HR at the GET with the standard first HRV threshold (HRVT1s, DFA a1 = 0.75), and a custom first HRV threshold (HRVT1c) based on the DFA a1 response observed early in the ramp. It is hypothesized that as in some prior studies, the conventional HRVT1s will overestimate the V̇O2/HR in relation to the GET but after correction of the individual midpoint, the HRVT1c would show good agreement with established standards. Additionally, since no change in HRVT2 methodology is occurring, there would still be comparable agreement to the RCP as seen in earlier studies.

RR Measurements, HRVT1s, HRVT1c, HRVT2
Each participant's RR time series was recorded by a Polar H10 strap (Polar Electro, Kempele, Finland). The Polar strap electrodes were covered with conductive gel and firmly fitted to the subpectoral area with the module centered over the sternum. Before testing, the Polar H10 ECG waveform was visually evaluated with the Android app ECG Logger (https://ecglogger.en.aptoide.com/app). To optimize DFA a1 measurements, the strap was shifted slightly to the left if the R peak amplitude was lower than the S wave3. H10 data was transmitted through Bluetooth to an Android smartphone using the ECG Logger app for recording both RR intervals, ECG and time alignment for further analysis.
Thresholds corresponding to HRVT1s, HRVT1c and HRVT2 were calculated by a third evaluator who was blinded to the gas exchange and ventilatory data results. RR data previously recorded for each participant were imported into Kubios Scientific HRV Software (Version 4.1, Biosignal Analysis and Medical Imaging Group, Department of Physics, University of Kuopio, Kuopio, Finland). Kubios preprocessing settings were set to a detrending method of “Smoothness priors”, with the smoothing parameter at 500, and the cutoff frequency at 0.035. DFA short-term fluctuation window width was set to 4 ≤ n ≤ 16 beats18. Visual inspection of the entire test recording was performed to determine sample quality, noise, arrhythmia, and missing beat artifact. The RR series of each participant was corrected by the Kubios “automatic method” and DFA a1 was calculated every 5 seconds with 2-minute measurement windows and results exported as text files for further analysis. Acceptable percent artifact occurring during threshold interpretation segments was set to below 5% 22.
Plotting of DFA a1 vs HR was performed for the determination of HRVT1s HR and HRVT2 HR as seen in Figure 1 and detailed previously18. HRVT1s HR defined as the HR where DFA a1 equaled 0.75 and HRVT2 HR defined as the HR where DFA a1 equaled 0.5 18.
The HRVT1s and HRVT2 V̇O2 was derived via the DFA a1 vs time plot. The times that DFA a1 reached 0.75 (HRVT1s) and 0.5 (HRVT2) were used to then derive the V̇O2 via the V̇O2 vs time relation based on gas exchange and ventilatory analysis 18.
The determination for HRVT1c HR and V̇O2 was identical to the above, except that instead of specifying DFA a1 = 0.75 as the first threshold, each participant was assigned a custom value. The HRVT1c custom value was defined as the midpoint between the maximal DFA a1 (Max DFA a1) value seen in the early portion of the ramp incremental and 0.5. This maximal value was always below 3 SD from the running 45 second local mean value. For example, if the maximal value during the early ramp incremental portion was 1.5, the corresponding V̇O2/HR HRVT1c was based on a DFA = 1.0 (midway between 1.5 and 0.5) as seen in Figure 1.

Results
Oxygen Uptake Agreement:
GET and HRVT1s: Group mean values were statistically different (p = 0.010, d = 0.80; Table 2, Figure 2). Bland Altman analysis showed lower GET than HRVT1s (bias = 4.4, LOA -6.3 to 15 mL·kg-1·min-1; Figure 3). Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4. As seen in Figure 5, the GET vs HRVT1s were not equivalent.
GET and HRVT1c: Group mean values were not statistically different (p = 0.77, d = 0.08; Table 2, Figure 2). Bland Altman analysis showed similar GET and HRVT1c (bias = -0.4, LOA -9.0 to 8.3 mL·kg-1·min-1; Figure 3). Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4. As seen in Figure 5, the GET vs HRVT1c were equivalent.
RCP and HRVT2: Group mean values were not statistically different (p = 0.26, d = 0.31; Table 2, Figure 2). Bland Altman analysis showed similar RCP and HRVT2 (bias = -1.4, LOA -10.2 to 7.4 mL·kg-1·min-1; Figure 3). Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4.
Heart Rate Agreement:
GET and HRVT1s: Group mean values were statistically different (p = 0.005, d = 0.91; Table 2, Figure 2). Bland Altman analysis showed lower GET than HRVT1s (bias = 16, LOA -18 to 50 bpm; Figure 3). Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4. As seen in Figure 5, the GET vs HRVT1s were not equivalent.
GET and HRVT1c: Group mean values were not statistically different (p = 0.65, d = 0.12; Table 2, Figure 2). Bland Altman analysis showed similar GET and HRVT1c (bias = 2, LOA -24 to 27 bpm; Figure 3. Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4. As seen in Figure 5, the GET vs HRVT1c were equivalent.
RCP and HRVT2: Group mean values were not statistically different (p = 0.98, d = 0.007; Table 2, Figure 2). Bland Altman analysis showed similar RCP and HRVT2 (bias = 0, LOA -21 to 21 bpm; Figure 3). Regression plots with Pearson’s r, SEE and ICC are presented in Figure 4.
Mean Max DFA a1 during the early ramp incremental was 1.52 ± 0.22 with mean calculated HRVT1c at a DFA a1 of 1.01±0.11.





Discussion
The purpose of this study was to explore the effects of modifying the value of DFA a1 used for determining the oxygen uptake and heart rate associated with the first HRV threshold, from a fixed value of 0.75 for all participants, to that of one based on individual DFA a1 ramp characteristics. Findings indicate that by using a custom first HRV threshold derived from the DFA a1 value midway between the maximal seen during the early portion of the ramp incremental and 0.5, the positive bias of HR and V̇O2 observed using the conventional HRVT1s was eliminated. Furthermore, since the line of regression used for determining HRVT1c was unchanged from standard methodology (Figure 1), the tight relationship between the HRVT2 and RCP was maintained.
HRVT1s vs HRVT1c agreement to the GET
As in some previous reports, the V̇O2/HR seen at the HRVT1s was substantially higher than the GET V̇O2/HR 12,14,15,17. This bias was virtually eliminated by individually defining each participant’s DFA a1 predicted to correspond to the GET, with no differences in group mean HRVT1c V̇O2/HR comparisons to GET V̇O2/HR. Additionally, there was a reduction in the LOA (for example LOA reduced from  a range of ±35 to ±26 bpm for HR), with trivial bias in the V̇O2/HR associated to HRVT1c compared to the V̇O2/HR responses associated to the HRVT1s (Figure 3). In terms of group correlation to the GET V̇O2/HR, both HRVT1s and HRVT1c showed similar strength in Pearson’s r and ICC to prior studies11,15,18,19,21 (Figure 4).  Equivalence between GET and HRVT1c V̇O2/HR was also verified (Figure 5).  Mean Max DFA a1 during the ramp incremental was 1.5, resulting in a calculated HRVT1c threshold near 1.0. These observations support the hypothesis that a given individual’s DFA a1 ramp incremental midpoint between well correlated and an uncorrelated pattern does not necessarily occur at a value of 0.75. In past studies, early ramp incremental DFA a1 values were not systematically provided. However, it is of interest to highlight a group of participants that showed both good agreement of the HRVT1s to the GET and also had DFA a1 recorded during a low intensity warmup pre-ramp 16,19. The mean DFA a1 during the pre-ramp warmup was 1.12±0.23 (corresponding HR = 141±13), making the calculated DFA a1 at HRVT1c of 0.81 and standard HRVT1s of 0.75 almost the same. Thus, as with the current report, matching the V̇O2/HR to the calculated DFA a1 ramp incremental midpoint resulted in good first HRV threshold to GET agreement via mathematical adjustment of first threshold targets. Furthermore, the high day-to-day reliability 16 of each participant’s warmup DFA a1 (ICC = 0.85) and the repeatability of the HRVT214,15,17 adds confidence to the notion that midpoint calculation might be a repeatable occurrence.


HRVT2 vs RCP agreement
There was good agreement between the RCP V̇O2/HR with that of the HRVT2 V̇O2/HR in terms of group mean values (Table 2), minimal bias, moderate LOA (Figure 3) and high degrees of correlation (Figure 4). This was not unexpected as the recording device, procedures, calculating software and plotting methodology were the same as in previous reports which indicated similar findings 11–14,21. However, it was important to present this data not only to corroborate previous results, but also to highlight that there is no alteration in the derivation of the HRVT2 as a significant advantage of the proposed HRVT1c modification. 


Contrasting ANS vs cardiorespiratory thresholds
While the use of a custom HRVT1c calculation did improve overall agreement with the GET, variation still exists on an individual basis. This may partially relate to the underlying conceptual differences between a marker of ANS response and measures derived from gas exchange indicators such as V̇O2, V̇CO2 and ventilatory responses which may diverge in some individuals. HRV in general arises from the competing influences of the stimulatory sympathetic and inhibitory parasympathetic branches of the ANS on the sinoatrial node resulting in fractal correlation behavior of the cardiac beat sequence 23,24. It is theorized that a potential rationale for this behavior is an attempt by the cardiovascular system to best cope with a rapidly changing environment 4,32,33. These environmental changes affect multiple neuromuscular, biochemical, peripheral and central nervous system inputs leading to alterations in “systems integration”34. Within this framework, various physiologic responses and cardiovascular advantages may exist behind the changes seen in correlation patterns due to increasing exercise intensity and overall organismic demands35. Therefore, at low to moderate exercise intensity where future physiologic requirements may be quite variable, cardiac ANS measures consistent with plasticity/flexibility are preferred (DFA a1 is correlated, 0.75 ≤ n ≤ 1.5) but these same measures becomes more rigid (DFA a1 is uncorrelated/anticorrelated, ≤ 0.5), in the heavy to severe domain as an ultimate protective response 32,36. Anticorrelated behavior refers to a pattern viewed as an immediate self-correction mechanism associated with the potential failure of homeostatic regulation and can only be tolerated for short time spans, in agreement with the severe intensity domain 26. Consistent with entering this metabolically unstable zone, it has also been shown that DFA a1 drops to below 0.5 in normal volunteers during high dose norepinephrine infusion 37. An analogy to these correlation behaviors could be optimization strategies employed by a soccer goalie. During most game play, the goalie would maintain a flexible approach, able to respond to highly variable threats (as with the HRVT1c). Under more critical circumstances, as in an imminent scoring opportunity, the choice of goalie position would be greatly constrained to prevent a catastrophic response (as with DFA a1 ≤ 0.5). The consequence of these conceptual differences in correlation properties (flexibility vs rigidity) is that the first HRV threshold may need to be mathematically realigned to account for individual variability in the DFA a1 ramp incremental range, as done in the present report. 


Experimental considerations
As previously discussed, many circumstances can alter DFA a1 determination and the resulting HRV thresholds including prior fatigue, software preprocessing (detrending), ECG/RR artifact correction, recording device bias and chest belt placement 3,22. Additionally, little data exists comparing ramp incremental derived HRVTs with that of DFA a1 during constant load exercise. It has been recognized that lengthy time spans at low intensity 38 or short time spans at high intensity 19 can suppress the DFA a1 that would otherwise be expected. As a consequence of the different underlying physiological concepts involving cardiorespiratory thresholds and markers of ANS status, individualistic differences may remain between the GET/RCP and ANS linked HRVTs even with further advances in HRV methodology. Continued research into combining other noninvasive surrogate markers with the HRVT such as muscle NIRS desaturation breakpoints 13, or ECG derived respiratory rate thresholds 12 appears reasonable despite the increased cost or complexity involved. Additionally, even established methods such as gas exchange based respiratory responses have both error, device bias and day to day variations to some degree39. With regards to equivalence testing, setting different boundaries for Cohen’s d could affect these results.


Practical Applications
The results of this study refine and support the original hypothesis that the V̇O2/HR found at the GET would correspond with that of the DFA a1 seen at a midpoint between well correlated and uncorrelated values. However, instead of a fixed midpoint value of 0.75 used for all circumstances, the novel finding presented here indicates that this target needs to be derived from the maximum DFA a1 seen early in an individual’s ramp incremental. Therefore, by simply altering a single facet of the first HRV threshold computation, we can resolve the differences between conflicting study outcomes such as the poor GET to HRVT1s agreement in this data set with that of the stronger agreement seen by Van Hooren et. al. 19. However, even with the HRVT1c adjustment, underlying differences between ANS vs cardiorespiratory responses may preclude exact agreement between HRV related thresholds and established standards.


Conclusion
An individualized first HRV threshold estimate based on the V̇O2/HR occurring at the DFA a1 midpoint between that observed early in the ramp incremental and an uncorrelated pattern showed greater agreement to the GET than the conventional methodology using a fixed DFA a1 of 0.75. Mean values for GET V̇O2/HR and that of HRVT1c derived values were in closer agreement than what was observed using the standard approach. Application of this refinement to existing methodology will hopefully resolve conflictive first HRV threshold study outcomes, while preserving the already established RCP to HRVT2 relationship.

Personal comments:

First off, many thanks to Juan and Pablo for making this happen. What an honor to work with one of the top exercise physiologists (and physiology labs) in the world. Additionally, these investigators are renown for their work on accurate gas exchange thresholds, making the collaboration even more appropriate. The HRV data was analyzed in a totally blind fashion by myself (with no knowledge of gas exchange data), making the study a strong one.

Is this the end of the HRVT1 controversy? I doubt it (see below). There will continue to be demographics ill-suited to ANS threshold surrogates (post covid comes to mind) and we are still in the early stages of understanding dynamic HRV pitfalls and usage. However, this mathematical correction seems to go a long way to reconciling the conflicts seen in the literature. The LOAs are still high, making a case for adding another surrogate marker (as in EDRT).

Although not focused on, the RCP to HRVT2 agreement continues to be consistent with prior work and appears to be a solid concept. 

A postscript...

You may or may not have noticed in our methods (in bold above) that we go to great lengths to optimize the H10 belt placement, to get the best R peak voltage possible. Why, it's simple - poor R peaks will skew the a1 and also lead to potential artifact (from poor signal-to-noise ratio). One can also get RR bias with alternate ECG leads, making placement an overlooked issue in a1 research.


This practice has been done with the Murias team related studies, with excellent results. As you can see in our article looking at SmO2 and HRVT2 combo thresholds, there were no cases of missed beat artifact (just two cases of arrhythmias that precluded analysis). The reason I mention this is a recent paper concluding a1 thresholds were not very aligned with gas exchange data. I looked over the methods, and on first perusal, found nothing wrong. Kubios - check, H10 - check, gas exchange - check. 

What did they find? That HRVT1 occurred at an average a1 of 0.67!!! I put the exclamation marks because no one has seen that to my knowledge. If anything, the a1 at the GET/VT1 has been above 0.75 (we have been seeing near 1.0).

The midpoint of a1 was not calculated, but it did not appear to be an issue based on a sample they showed:


 

So what are some factors that lead to early a1 decline?

  • Fatigue, recent HIT - no mention of abstaining from recent exercise was in the methods. It is possible there was an effect from fatigue/recent training/racing.
  • Poor R peaks - as noted above, poor signal to noise ECG signals can lead to both artifacts (which the authors were complaining about) and an effect on the RR precision, causing a1 to measured lower than it should. this is my best guess.
  • Older generation HRMs - the H7 had this issue

What do other groups see regarding average a1 at the VT/GET? A recent study, mirrored ours closely in that the HRVT1 HR was well above the GET HR, which is what we saw in the MSSE publication. 

Is this paper a case of garbage in, garbage out - I just don't know? The authors make no attempt to scientifically explain the discrepancy, just disparage the process. Next time they work with the H10, perhaps they can read the instructions...

 

 

References

1.   Meyer T, Lucía A, Earnest CP, Kindermann W. A conceptual framework for performance diagnosis and training prescription from submaximal gas exchange parameters--theory and application. Int J Sports Med. 2005;26 Suppl 1:S38-48. doi:10.1055/s-2004-830514
2.    Keir DA, Fontana FY, Robertson TC, et al. Exercise Intensity Thresholds: Identifying the Boundaries of Sustainable Performance. Med Sci Sports Exerc. 2015;47(9):1932-1940. doi:10.1249/MSS.0000000000000613
3.    Rogers B, Gronwald T. Fractal Correlation Properties of Heart Rate Variability as a Biomarker for Intensity Distribution and Training Prescription in Endurance Exercise: An Update. Front Physiol. 2022;13:879071. doi:10.3389/fphys.2022.879071
4.    Goldberger AL. Non-linear dynamics for clinicians: chaos theory, fractals, and complexity at the bedside. Lancet Lond Engl. 1996;347(9011):1312-1314. doi:10.1016/s0140-6736(96)90948-4
5.    Hardstone R, Poil SS, Schiavone G, et al. Detrended Fluctuation Analysis: A Scale-Free View on Neuronal Oscillations. Front Physiol. 2012;3. doi:10.3389/fphys.2012.00450
6.    Goldberger AL, Amaral LAN, Hausdorff JM, Ivanov PCh, Peng CK, Stanley HE. Fractal dynamics in physiology: Alterations with disease and aging. Proc Natl Acad Sci. 2002;99(suppl_1):2466-2472. doi:10.1073/pnas.012579499
7.    Gronwald T, Rogers B, Hoos O. Fractal Correlation Properties of Heart Rate Variability: A New Biomarker for Intensity Distribution in Endurance Exercise and Training Prescription? Front Physiol. 2020;11:550572. doi:10.3389/fphys.2020.550572
8.    Karapetian G, Engels H, Gretebeck R. Use of Heart Rate Variability to Estimate LT and VT. Int J Sports Med. 2008;29(08):652-657. doi:10.1055/s-2007-989423
9.    Blasco-Lafarga C, Camarena B, Mateo-March M. Cardiovascular and Autonomic Responses to a Maximal Exercise Test in Elite Youngsters. Int J Sports Med. 2017;38(09):666-674. doi:10.1055/s-0043-110680
10.    Rogers B, Giles D, Draper N, Mourot L, Gronwald T. Detection of the Anaerobic Threshold in Endurance Sports: Validation of a New Method Using Correlation Properties of Heart Rate Variability. J Funct Morphol Kinesiol. 2021;6(2):38. doi:10.3390/jfmk6020038
11.    Mateo‐March M, Moya‐Ramón M, Javaloyes A, Sánchez‐Muñoz C, Clemente‐Suárez VJ. Validity of detrended fluctuation analysis of heart rate variability to determine intensity thresholds in elite cyclists. Eur J Sport Sci. 2023;23(4):580-587. doi:10.1080/17461391.2022.2047228
12.    Rogers B, Schaffarczyk M, Gronwald T. Improved Estimation of Exercise Intensity Thresholds by Combining Dual Non-Invasive Biomarker Concepts: Correlation Properties of Heart Rate Variability and Respiratory Frequency. Sensors. 2023;23(4):1973. doi:10.3390/s23041973
13.    Fleitas-Paniagua PR, de Almeida Azevedo R, Trpcic M, Murias JM, Rogers B. Combining Near-Infrared Spectroscopy and Heart Rate Variability Derived Thresholds to Estimate the Critical Intensity of Exercise. J Strength Cond Res. 2024;38(1):e16. doi:10.1519/JSC.0000000000004597
14.    Fleitas-Paniagua PR, Marinari G, Rasica L, Rogers B, Murias JM. Heart Rate Variability Thresholds: Agreement with Established Approaches and Reproducibility in Trained Females and Males. Med Sci Sports Exerc. Published online February 12, 2024. doi:10.1249/MSS.0000000000003412
15.    Sempere-Ruiz N, Sarabia JM, Baladzhaeva S, Moya-Ramón M. Reliability and validity of a non-linear index of heart rate variability to determine intensity thresholds. Front Physiol. 2024;15. Accessed March 6, 2024. https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2024.1329360
16.    Van Hooren B, Bongers BC, Rogers B, Gronwald T. The Between-Day Reliability of Correlation Properties of Heart Rate Variability During Running. Appl Psychophysiol Biofeedback. 2023;48(4):453-460. doi:10.1007/s10484-023-09599-x
17.    Fleitas‐Paniagua PR, De Almeida Azevedo R, Trpcic M, Murias JM, Rogers B. Effect of ramp slope on intensity thresholds based on correlation properties of heart rate variability during cycling. Physiol Rep. 2023;11(15):e15782. doi:10.14814/phy2.15782
18.    Rogers B, Giles D, Draper N, Hoos O, Gronwald T. A New Detection Method Defining the Aerobic Threshold for Endurance Exercise and Training Prescription Based on Fractal Correlation Properties of Heart Rate Variability. Front Physiol. 2021;11:596567. doi:10.3389/fphys.2020.596567
19.    Van Hooren B, Mennen B, Gronwald T, Bongers BC, Rogers B. Correlation properties of heart rate variability to assess the first ventilatory threshold and fatigue in runners. J Sports Sci. 2023;0(0):1-10. doi:10.1080/02640414.2023.2277034
20.    Rogers B, Berk S, Gronwald T. An Index of Non-Linear HRV as a Proxy of the Aerobic Threshold Based on Blood Lactate Concentration in Elite Triathletes. Sports. 2022;10(2):25. doi:10.3390/sports10020025
21.    Schaffarczyk M, Rogers B, Reer R, Gronwald T. Validation of a non-linear index of heart rate variability to determine aerobic and anaerobic thresholds during incremental cycling exercise in women. Eur J Appl Physiol. 2023;123(2):299-309. doi:10.1007/s00421-022-05050-x
22.    Rogers B, Giles D, Draper N, Mourot L, Gronwald T. Influence of Artefact Correction and Recording Device Type on the Practical Application of a Non-Linear Heart Rate Variability Biomarker for Aerobic Threshold Determination. Sensors. 2021;21(3):821. doi:10.3390/s21030821
23.    Goldberger A. Is the Normal Heartbeat Chaotic or Homeostatic? Physiology. 1991;6(2):87-91. doi:10.1152/physiologyonline.1991.6.2.87
24.    White DW, Raven PB. Autonomic neural control of heart rate during dynamic exercise: revisited. J Physiol. 2014;592(12):2491-2500. doi:10.1113/jphysiol.2014.271858
25.    Gronwald T, Hoos O. Correlation properties of heart rate variability during endurance exercise: A systematic review. Ann Noninvasive Electrocardiol. 2020;25(1):e12697. doi:10.1111/anec.12697
26.    Seely AJ, Macklem PT. Complex systems and the technology of variability analysis. Crit Care. 2004;8(6):R367. doi:10.1186/cc2948
27.    Gronwald T, Hoos O, Ludyga S, Hottenrott K. Non-linear dynamics of heart rate variability during incremental cycling exercise. Res Sports Med Print. 2019;27(1):88-98. doi:10.1080/15438627.2018.1502182
28.    Iannetta D, Murias JM, Keir DA. A Simple Method to Quantify the V˙O2 Mean Response Time of Ramp-Incremental Exercise. Med Sci Sports Exerc. 2019;51(5):1080-1086. doi:10.1249/MSS.0000000000001880
29.    Keir DA, Iannetta D, Mattioni Maturana F, Kowalchuk JM, Murias JM. Identification of Non-Invasive Exercise Thresholds: Methods, Strategies, and an Online App. Sports Med Auckl NZ. 2022;52(2):237-255. doi:10.1007/s40279-021-01581-z
30.    Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res. 1999;8(2):135-160. doi:10.1177/096228029900800204
31.    Lakens D. Equivalence Tests. Soc Psychol Personal Sci. 2017;8(4):355-362. doi:10.1177/1948550617697177
32.    Fossion R, Rivera AL, Estañol B. A physicist’s view of homeostasis: how time series of continuous monitoring reflect the function of physiological variables in regulatory mechanisms. Physiol Meas. 2018;39(8):084007. doi:10.1088/1361-6579/aad8db
33.    Ivanov PCh  null, Nunes Amaral LA, Goldberger AL, Stanley HE. Stochastic feedback and the regulation of biological rhythms. Europhys Lett. 1998;43(4):363-368. doi:10.1209/epl/i1998-00366-3
34.    Balagué N, Torrents C, Hristovski R, Kelso J a. S. Sport science integration: An evolutionary synthesis. Eur J Sport Sci. 2017;17(1):51-62. doi:10.1080/17461391.2016.1198422
35.    Balagué N, Hristovski R, Almarcha M, Garcia-Retortillo S, Ivanov PCh. Network Physiology of Exercise: Vision and Perspectives. Front Physiol. 2020;11:611550. doi:10.3389/fphys.2020.611550
36.    Muñoz MA. Colloquium : Criticality and dynamical scaling in living systems. Rev Mod Phys. 2018;90(3):031001. doi:10.1103/RevModPhys.90.031001
37.    Tulppo MP, Mäkikallio TH, Seppänen T, et al. Effects of pharmacological adrenergic and vagal modulation on fractal heart rate dynamics. Clin Physiol. 2001;21(5):515-523. doi:10.1046/j.1365-2281.2001.00344.x
38.    Rogers B, Mourot L, Doucende G, Gronwald T. Fractal correlation properties of heart rate variability as a biomarker of endurance exercise fatigue in ultramarathon runners. Physiol Rep. 2021;9(14). doi:10.14814/phy2.14956
39.    Van Hooren B, Souren T, Bongers BC. Accuracy of respiratory gas variables, substrate, and energy use from 15 CPET systems during simulated and human exercise. Scand J Med Sci Sports. 2024;34(1):e14490. doi:10.1111/sms.14490



Monday, September 23, 2024

Garmin Resting HRV - my experience

Garmin Resting HRV - my experience

As the basis for many fitness related concepts, the measurement of resting HRV is of critical importance. These metrics include "readiness to train", sleep quality/stages and autonomic strain/recovery. Consequently, even if some of the newer "hyped" tracking devices have these features, we are still at the mercy of whether they can accurately measure resting HRV (yes I'm talking about the Fenix 8 and the newer "rings"). Over the past several years, I have used numerous Garmin devices, including the Fenix 5, 6, 7, Epix 2 and FR965. Despite generational advances in the optical sensors (according to the manufacturer), my experience has not been positive regarding resting HRV (or exercise HR). Time after time, my H10 or Movesense ECG overnight readings (using Kubios with an ECG) were 200-300% different from Garmin! This frustrated me to no end considering my interest in accurate HRV.....So, I sought a solution to the problem.

 

Let's go through some of my testing and a solution to this issue. 

Baseline - A typical night recording with my Garmin FR 965:

  • Garmin FR 965 (wrist) RMSSD = 47 ms:

 

Comparison to Gold standard 

And at the same time wearing a Movesense ECG, sample rate 250 Hz with Kubios HRV results: 

  • Movesense ECG with Kubios, full night average RMSSD = 18 ms:

  • The Garmin unit is about 2.5x too high!
  • I've done this many times before on all the Garmin devices mentioned above, and if anything, the error is getting worse with the newer units. You can see my Garmin range of 46 to 56 ms, which I never see on ECG overnight measurements.
  • Hence, the Garmin HRV related metrics/recommendations/sleep phases are assumed to be inaccurate. 

My solution:

My wife has an Apple Watch and from my testing (as above) the HRV and exercise HR have been spot on! But I'm an android user, so that's out for me. My thought was that perhaps the big "players" like Apple (who could attract the best developers through higher budgets and benefits) might provide the best accuracy, so I did the next best thing - I bought a Pixel 3 watch (Google/Fitbit). 

Here are the initial HRV results (I've only had it a week or so).

Gold standard -  

  • Movesense ECG with Kubios, full night average RMSSD = 14.3ms:


 

  • Pixel Watch 3 wrist (RMSSD = 14 ms):

  • As you can see with the Garmin (46-50ms) and Pixel (14-18ms) daily RMSSD numbers, they are internally consistent within device, with only small changes on a day-to-day basis. 

What surprises me the most is not the huge error in the Garmin metric (I already knew that for years), but how close the Pixel was aligned with Kubios/ECG results. This is very encouraging, and I'll be keeping track over the next few weeks to make sure the good agreement continues.

  • What does this mean for those of you using various rings and watches? It means that your device may or may not be in alignment with true resting HRV. If it is, then great, but if not, then the "recommendations" put forth by (let's say) Garmin may be fatally flawed.

The solution

There is a simple way of knowing where you and your device stands - use the Polar H10 chest belt, record a full night of data (with Fatmaxxer, Garmin units or other apps) and use Kubios (free version) to ascertain the RMSSD. You may want to temporarily disable the Garmin from its pairing with the H10 just to be sure that it's using optical input.

In this post I have a brief tutorial on how to use Kubios free version (you will need to use a measurement window size of the entire night, not 2 minutes as used for DFA a1). If this is too complex, there is a simpler way - use Fatmaxxer, which does log the RMSSD to the "features" file. Yet again, another great feature of Fatmaxxer! These would be 2-minute window measurements, so you will need to average the column:

This is the "features" file from Kubios (not mine, a random test case):

 

My night data test: Full night measurement window vs averaging all the 2 minute Fatmaxxer measurements:

  • I did a comparison between the RMSSD of an 8 hour nightly session as a single measurement window in Kubios (8 hrs) vs the same RMSSD as 2 minute windows averaged (over 6000 overlapping Kubios samples). The results were quite close - 17 vs 18 ms.

 

Exercise HR - Outdoor road cycling

The other item I was curious about was how well the Pixel 3 watch optical HRM agreed with the H10 in simple HR. In the past, the Garmin devices have been dismal for outdoor cycling and even unreliable indoors on a trainer (for me). Therefore, I did about 3 hours outdoors on bumpy roads with a few intervals. Here is a partial sample of the results:

  • You get the picture - and it's impressive for a wrist device during road riding at various intensities.

Why should we care about absolute values?

You may think that it's ok to have this systematic huge bias in HRV as long as the trend is helpful. Perhaps. But there is a great deal of value in knowing your true resting RMSSD (or SDNN). Here are a couple of interesting studies looking at mortality prediction and biologic aging


 


Summary and Thoughts

  • The above results are based on an N=1 sample, me. Perhaps your device/wrist combo will function differently and match to a gold standard. However, I do want to caution getting a false sense of confidence after a single good correlation between a Kubios or Fatmaxxer RMSSD sample and your device. What if "X" vendor's RMSSD (Garmin, Suunto, etc.) just coincidently matches your reading with Kubios? After all, my Garmin readings are certainly plausible for many individuals and could match gold standards just from statistical luck. How do we get around this circumstance? In that case, one could follow along for several nights and make sure the correlation holds up.
  • Further, this type of discrepancy in a basic wearable metric vs an established method could be present elsewhere in a company's product. If Garmin can't get optical HRV correct, what else are they messing up that we can't easily measure? This illustrates the need for us to "trust but verify". It's unfortunate we can't blindly trust what the various wearable vendors are saying, but these are not FDA approved medical devices (therefore, anything goes). 
  • The other clue to optical HRM quality could also be how well the HR tracks during exercise. If your wearable aligns well with a chest belt, especially during intervals, that could be a favorable sign. 
  • I have abandoned my Garmin FR965 after being a loyal user for many years. I still use the Edge 1050 for cycling (which is great for recording anything ANT+ or bluetooth and displaying alphaHRV a1). I'd rather have accurate resting HRV (and wrist exercise HR) than the nice watch faces.
  • Bottom line - Resting HRV is a promising metric for readiness, ANS stress, illness and recovery. Please make sure your readings are trustworthy....